Open Access
Add to BibSonomyAbstract
In this work, we present a comparative study of two just-in-time (JIT) dynamic bandwidth allocation algorithms (DBAs), designed to improve the energy-efficiency of the 10 Gbps Ethernet passive optical networks (10G-EPONs). The algorithms, termed just-in-time with varying polling cycle times (JIT) and just-in-time with fixed polling cycle times (J-FIT), are designed to achieve energy-savings when the idle time of an optical network unit (ONU) is less than the sleep-to-active transition time. This is made possible by a vertical-cavity surface-emitting laser (VCSEL) ONU that can transit into sleep or doze modes during its idle time. We evaluate the performance of the algorithms in terms of polling cycle time, power consumption, percentage of energy-savings, and average delay. The energy-efficiency of a VCSEL ONU that can transition into sleep or doze mode is compared to an always-ON distributed feedback (DFB) laser ONU. Simulation results indicate that both JIT and J-FIT DBA algorithms result in improved energy-efficiency whilst J-FIT performs better in terms of energy-savings at low network loads. The J-FIT DBA however, results in increased average delay in comparison to the JIT DBA. Nonetheless, this increase in average delay is within the acceptable range to support the quality of service (QoS) requirements of the next-generation access networks.
© 2013 Optical Society of America
1. Introduction
In recent years, we have witnessed a substantial growth in the Internet traffic as a result of bandwidth-intensive services like Video-on-demand (VoD) and IP telephony. To cater for this increase in Internet traffic, the passive optical network (PON) has been identified as a favourable access network architecture [1, 2]. The major contribution towards the energy consumption in the PON is attributed to the optical network units (ONUs) [3, 4], placed at the customer premises. As a result of the broadcast nature of the downstream traffic in PONs, each ONU receives packets destined for every ONU in the PON. The ONU receiver processes these packets before selecting the ones destined to them and discarding the rest. The ONU receiver therefore consumes a large percentage of energy in processing packets not destined to it [5]. To minimize this unwanted energy consumption at the ONU, the ITU-T has proposed energy-saving techniques such as sleep and doze modes [6]. In sleep mode, both the ONU transmitter and the receiver are powered down in the absence of traffic. In doze mode however, only the ONU transmitter is powered down, facilitating the ONUs to remain in sync with the optical line terminal (OLT).
The energy consumption of the ONU depends on the choice of the dynamic bandwidth allocation (DBA) algorithm used in the PON. This has motivated many energy-efficient DBA algorithms incorporating sleep and doze modes in PON. Smith et. al[7] proposed a bandwidth allocation algorithm to frequently transit the ONUs into sleep mode in Gigabit Passive Optical Networks (GPONs). This algorithm assigns a minimum bandwidth in the upstream for ONUs entering sleep mode and facilitates ONUs to send their bandwidth requests when upstream traffic is present at the ONU. Kubo et. al[8] discussed how energy consumption can be reduced by using a hybrid mechanism that combines sleep mode and adaptive link rate control (LRC). The sleep mode mechanism switches the ONU between sleep and active modes depending on the presence or absence of traffic. The adaptive LRC mechanism switches the link rate between 1 Gbps and 10 Gbps depending on the amount of traffic, saving energy even when there is traffic to be transmitted. Ren et. al[9] proposed a mathematical approach to analyze the 10 Gbps PON using Poisson distributed traffic. Wong et. al[10] proposed a Just-In-Time (JIT) bandwidth allocation algorithm for Time Division Multiplexing (TDM) PON using an ONU with fast wake up capability. This JIT algorithm allows ONUs to go into variable sleep lengths based on a sleep time optimiser and uses the same GATE and REPORT messages as in IEEE 802.3ah standard [11] to transition the ONU to and from sleep mode. Fiammengo et. al[12] proposed a criteria for determining the sleep time of an ONU based on the inter-arrival time of incoming traffic at the ONU. Praet et. al[5] proposed a bit-interleaving TDM downstream protocol (Bi-PON) to minimize the power consumed in complex processing of packets, destined for other ONUs in the PON. The protocol organizes the bits, intended for a certain ONU, in an interleaved form thereby reducing the number of functional blocks required to extract packets at the ONU. The Bi-PON protocol also incorporates sleep command in the operation, administration and maintenance (OAM) header to further improve the energy-savings during the idle time.
To date existing research has only assessed the possible energy-savings that could be achieved with the sleep mode capability of an ONU. In previous works, when the ONU idle time ≤ sleep-to-active transition time, the ONUs remain in active mode, thus achieving no energy-savings. Therefore, the energy-efficiency of the PONs could be further improved if the ONU power consumption could be reduced for the ONU idle times ≤ sleep-to-active transition time. A DBA algorithm that can transition an ONU into doze mode as well as into sleep mode during its idle time would enhance the energy-efficiency of the PON further. For this purpose, we propose two DBA algorithms that accommodate an ONU that is capable of transitioning between active, sleep and doze modes for improved energy-savings. Using the doze capability of the ONU, the power consumption at the ONU can be further reduced when the ONU idle time is ≤ sleep-to-active transition time and ≥ doze-to-active transition time. To this end, we consider for illustrative purposes a 10 Gbps Ethernet passive optical network (10G-EPON) whereby the ONUs are able to sleep and doze (i.e. sleep/doze ONUs). The two algorithms, namely just-in-time DBA with varying polling cycle times (JIT) [13,14] and just-in-time DBA with fixed polling cycle time (J-FIT) are designed to exploit the vertical-cavity surface-emitting laser (VCSEL) ONUs that are capable of transitioning from doze-to-active mode in 330 ns and from sleep-to-active mode in 2 ms [15,16]. Both algorithms comply to the Multi-Point Control Protocol (MPCP) whereby GATE and REPORT control messages contain additional information for transitioning the VCSEL ONU between active, doze and sleep modes. Our JIT DBA algorithm transits the ONUs from sleep or doze mode into active mode just-in-time to receive data from the OLT. In this work, we show that the JIT DBA algorithm achieves tangible energy-savings even at low network loads whilst maintaining low delay at high network loads. Further, we propose another algorithm, namely the J-FIT DBA algorithm that uses fixed polling cycle times to extend the time an ONU stays in sleep or doze mode for improved energy-efficiency. In this work, we critically analyse and compare the performance of the two algorithms in terms of energy-efficiency and QoS parameters. We show that while both JIT and J-FIT result in improved energy-savings, the J-FIT DBA outperforms the JIT DBA, especially at low network loads. Our findings indicate that when compared to the JIT DBA, the J-FIT DBA results in an increase in average delay but the delay values still support the QoS requirements of the network. In a further analysis on the energy-efficiency of the VCSEL ONU using the JIT and J-FIT DBA algorithms, we show that the VCSEL ONU outperforms the distributed feedback (DFB) laser ONU.
The rest of the paper is structured as follows. In Section 2, we describe the JIT and J-FIT DBA algorithms in detail. Simulation parameters and results from the performance evaluation study of polling cycle time, power consumption, percentage of energy-savings, and average delay, are presented in Section 3. A summary of the paper is presented in Section 4.
2. Proposed DBA algorithms
An effective sleep/doze controlled just-in-time DBA algorithm is a one that can cater for varying traffic demands of different ONUs and can transit the ONUs from sleep or doze mode into active mode just-in-time to send and receive traffic. The remainder of this section describes how these objectives are achieved through our proposed JIT and J-FIT DBA algorithms. Applications such as peer-to-peer file and video sharing, Internet gaming, and video conferencing consume a significant bandwidth in the upstream direction. Based on the CISCO VNI forecast [17], a rapid growth is expected in these applications in the coming years. Considering this increase in upstream traffic, we have assumed symmetric upstream and downstream bandwidth in the network [10].
2.1. Just-In-Time DBA with varying polling cycle times
The actual traffic flow in the 10G-EPON as per the JIT DBA is shown in Fig. 1. The parameters Tcycle, Tsleep/doze, and Twakeup represent the polling cycle time, ONU sleep/doze time, and ONU wake up time, respectively. The data frames from the OLT to each ONU consist of data and the GATE message. When an ONU receives this data frame, it analyzes the GATE message to determine the transmission start time, transmission duration, and the sleep duration. The JIT DBA algorithm is designed such that the ONU starts data transmission to the OLT immediately after it completes analyzing the GATE message. When the ONU completes its upstream data transmission for a period specified in the GATE, the ONU generates the REPORT message and sends it to the OLT. The ONU then enters into sleep or doze mode for a period specified in the GATE and wakes up just-in-time to receive data from the OLT in the next cycle.
Figure 2 illustrates the flow chart of the JIT DBA executed at the OLT. In the JIT DBA, initially each ONU inspects its incoming queue and generates a REPORT message that contains the bandwidth requirement of that ONU. The OLT waits until it receives REPORT messages from all the ONUs. When all REPORT messages are received, the OLT processes all bandwidth requests and calculates the average bandwidth requested by each ONU. The OLT then compares the average bandwidth with the maximum allowable bandwidth that satisfies a given maximum polling cycle time. The maximum polling cycle time corresponds to the maximum time gap between two consecutive transmissions to a given ONU. This value is controlled by restricting the maximum allowable bandwidth per ONU to satisfy the delay and jitter of the network. Depending on the decision arising from this comparison, the ONUs are allocated either the maximum bandwidth or the average requested bandwidth in the subsequent polling cycle.
In any given cycle, the OLT has the knowledge of the bandwidth allocated to each ONU in advance. Thus, the OLT can determine the idle period of each ONU during each cycle. As we are considering that the 10G-EPON uses a 10 Gbps VCSEL ONU that can transit from doze-to-active mode in 330 ns, if the ONU idle time ≥ 2 ms, the OLT will instruct the ONU to enter sleep mode. Likewise, if 330 ns ≤ ONU idle time < 2 ms, the OLT will instruct the ONU to enter doze mode. Table 1 lists the transition times and power consumption values of the 10 Gbps VCSEL ONU considered in our work.
Table 1. Power consumption and switching values of VCSEL ONUs
Figure 3 shows the flow chart of the algorithm executed at each ONU. In any given cycle, an ONU wakes up from sleep or doze mode just-in-time to receive the GATE message and data from the OLT. The GATE message includes information on bandwidth allocated to the ONU. Apart from this traditional information about bandwidth allocation, the GATE message used in the JIT DBA algorithm includes the sleep start time, the sleep duration, and also the sleep/doze/active command. This is similar to the GATE message used in [10] but because we are introducing doze mode, the sleep/doze/active command is included the GATE message as well. After processing the GATE message, the ONU transmits data and then immediately enters into sleep or doze mode for a duration specified in the GATE message.
2.2. Just-In-Time DBA with fixed polling cycle times
In the JIT DBA algorithm, the polling cycle time corresponds to the sum of time slots allocated to each ONU. In turn, a time slot corresponds to the average requested bandwidth of the 10G-EPON. When the network load is low, the average requested bandwidth is low, resulting in shorter time slots. Thus, the sum of such short time slots results in shorter polling cycle times. Shorter polling cycle times thereby allow the ONUs to remain idle only for a shorter period. Short idle periods prevent the ONUs from going into sleep mode and instead the ONUs remain in active mode or enter into doze mode. Further, the difference in power consumption between doze and active modes of the VCSEL ONU is only 0.134 W (refer to Table 1) and therefore, this small difference in power consumption leads to low energy-savings at the ONUs at low network loads. Therefore, we propose the J-FIT DBA algorithm to overcome the shortfall of short polling cycle times of the JIT DBA through implementing fixed polling cycle times. The J-FIT DBA follows the argument that at low network loads, the ONUs can afford to sleep or doze for a longer duration without a significant compromise in QoS requirements such as delay.
The overall traffic flow in the 10G-EPON between the OLT and two ONUs using the J-FIT DBA is presented in Fig. 4. The J-FIT DBA is designed with a fixed polling cycle time that is independent of the average requested bandwidth of the 10G-EPON. The parameters Tfixed_cycle and Tidle represent the fixed polling cycle time and the ONU idle time, respectively. Similar to the JIT DBA algorithm, an ONU initially inspects its incoming queue and determines the bandwidth requirement in the next cycle. The ONU then generates the REPORT message and sends it to the OLT. Algorithm 1 presents a summary of the algorithm executed at the OLT after receiving REPORT messages from all the ONUs. The parameters Ttx_slot, Ttx_start, and Tmode represent the ONU transmission time, transmission start time, and ONU sleep/doze/active time, respectively. After receiving the REPORT messages, the OLT calculates the average bandwidth requested by each ONU. The OLT then compares this value against a threshold to ensure that for any given polling cycle, the total allocated bandwidth does not exceed the bandwidth that corresponds to Tfixed_cycle. Based on this comparison, the OLT assigns either the average requested bandwidth or maximum bandwidth to an ONU in the upcoming fixed polling cycle.
Algorithm 1. Pseudocode of the J-FIT DBA algorithm executed at the OLT for N number of ONUs
Figure 5 compares the polling cycle times of the JIT and J-FIT DBAs at ONU1, corresponding to two network loads L1 and L2 where L1 < L2, respectively. Referring to Fig. 5, in the J-FIT DBA, the Tidle of an ONU in a given cycle corresponds to Tfixed_cycle − Ttx_slot. As shown in Algorithm 1, the OLT determines the mode and the sleep/doze/active mode duration based on Tidle. This information is included in the GATE message destined to each ONU in the subsequent cycle.
In any given cycle, an ONU wakes up just-in-time to receive packets consisting of the GATE message and downstream data from the OLT. The ONU processes the GATE message and learns its upstream transmission start time and duration. After completing its upstream data transmission, the ONU generates the REPORT message that contains its future bandwidth requirement and sends it to the OLT. The ONU then enters into sleep, doze, or active modes for a duration specified in the GATE message. The J-FIT DBA is also designed to ensure that the upstream and downstream transmissions coincide under the assumption of equal bandwidth in both directions [10].
3. Simulations and results
The simulations for the JIT and J-FIT DBAs were performed using C++. Traffic arrivals at each ONU were generated according to a Poisson distribution. Table 2 lists the network and protocol parameters used in our simulations. Table 1 lists the power consumption and switching values of the DFB ONU used in comparing the energy-efficiency of the VCSEL ONU and DFB ONU under both JIT and J-FIT DBA algorithms.
Table 2. Network and Protocol Parameters
3.1. Performance comparison between JIT and J-FIT DBA algorithms
Figures 6(a) and 6(b) illustrate the average polling cycle time of the sleep/doze mode VCSEL ONU as a function of normalized network load and maximum and fixed polling cycle times of the JIT and J-FIT DBA algorithms, respectively. The normalized network load (eg. a normalized network load of 1 = 10 Gbps network load) is varied from 0.1 to 1. In the JIT DBA, for a given maximum polling cycle time, the average polling cycle time gradually increases with the increase in network load. When the network load increases, the number of packets arriving at the ONUs increases, thereby increasing the amount of bandwidth requested by each ONU. This increases the average bandwidth requested in the PON, thereby increasing the time slot allocated per ONU. As discussed in Section 2, the polling cycle time of the JIT DBA is the sum of time slots allocated per each ONU. Thus, an increase in the allocated time slot results in an increase in the polling cycle time. The J-FIT DBA however, is designed with fixed polling cycle times. Therefore, irrespective of the network load, the polling cycle time remains constant.
It is important to note that in Fig. 6(a), for the maximum polling cycle times of 2 ms and 4 ms, the average polling cycle time gradually increases with network load up to a load of 0.9 and then decreases for network loads ≥ 0.9. At the maximum polling cycle times 2 ms and 4 ms, as a result of the exponentially distributed traffic generated at each ONU, the maximum polling cycle time is reached between the network loads of 0.9 and 1. Therefore, the network load of 1 is not achievable under these two instances. This has resulted in the abnormal behaviour of the average polling cycle time at the maximum polling cycle times of 2 ms and 4 ms. Thus we eliminate these two points in our analysis.
Figure 7(a) illustrates the average power consumption per VCSEL ONU per polling cycle time as a function of normalized network load and maximum polling cycle time for the JIT DBA. The VCSEL ONU is capable of transitioning between active, doze and sleep modes. For the JIT DBA algorithm, for any given maximum polling cycle time, the average power consumption remains constant at 3.85 W up to a network load of 0.8. The average power consumption then decreases with the increase in network load. In the JIT DBA algorithm, the idle time of the ONUs remains below 2 ms up to a network load of 0.8 because of the low average polling cycle times shown in Fig. 6(a). This prevents an ONU from entering into sleep mode. As the idle time is higher than the doze-to-active transition time of 330 ns, the ONUs are transitioned into doze mode during its idle time. This has resulted in an average power consumption of 3.85 W, which corresponds to the power consumption of a VCSEL ONU in doze mode as given in Table 1. For network loads ≥ 0.8, the idle time satisfies the criteria for the ONUs to enter into sleep mode. As the power consumption of a sleep mode VCSEL ONU is 0.75 W, average power consumption values as low as 1.5 W could be achieved at network loads ≥ 0.8, for the JIT DBA.
Fig. 7 Average power consumption per VCSEL ONU per polling cycle time as a function of normalized network load and polling cycle time.
Figure 7(b) illustrates the average power consumption per VCSEL ONU per polling cycle time as a function of normalized network load and fixed polling cycle time for the J-FIT DBA. For the J-FIT DBA, for a given network load, the power consumption per ONU per fixed polling cycle time decreases with the increase in fixed polling cycle time. For a given network load, an increase in fixed polling cycle time allows an ONU to remain in sleep or doze mode for a longer period, which results in less power consumption per ONU. It is important to note that in the J-FIT DBA, the ONUs enter into sleep mode for fixed polling cycle times ≥ 2ms for any given network load. As the power consumption in sleep mode is 0.75 W, we can observe average power consumption values as low as 1.4 W at low network loads. For fixed polling cycle times ≤ 2ms however, the ONU enters into doze mode, resulting in an average power consumption of 3.85 W, which corresponds to the doze mode power consumption of a VCSEL ONU.
As shown in Fig. 7(b), for any given fixed polling cycle time, the power consumption increases with the increase in network load for the J-FIT DBA. When the network load increases, it increases the transmission time slots, i.e. the ONUs stays in active mode for a longer duration of time. The increased duration of time in active mode results in increased power consumption. Based on these results, both JIT and J-FIT DAB algorithms have resulted in improved energy-efficiency. The J-FIT however, performs better than JIT at low network loads in terms of energy-efficiency.
Figure 8(a) illustrates the percentage of energy-savings of a sleep/doze VCSEL ONU as compared to an always-ON VCSEL ONU, and as a function of network load and maximum polling cycle time for the JIT DBA. The percentage of energy-savings η1 is calculated as follows:
where the parameters Pvcsel,act and Pvcsel,doze/sleep represent the power consumption of a VC-SEL ONU in active mode and in sleep or doze modes, respectively. The parameters Tact and Tdoze/sleep represent the time an ONU spends in active and sleep or doze modes, respectively. As shown in Fig. 8(a), for any given maximum polling cycle time, η1 is 3% for network loads up to 0.8. For network loads ≥ 0.8, η1 increases with the increase in network load. As explained earlier, the VCSEL ONU enters into doze mode for network loads up to 0.8. As a result of the small difference in the power consumption (0.134 W) between active and doze modes of a VCSEL ONU, the η1 is small as well. For network loads ≥ 0.8, the ONU enters into sleep mode, resulting in energy-savings as high as 61%, due to the large difference in the power consumption (3.234 W) between active and sleep modes of a VCSEL ONU.Fig. 8 Percentage of energy-savings (η1) as a function of normalized network load and polling cycle time.
Figure 8(b) illustrates η1, as a function of network load and fixed polling cycle time for the J-FIT DBA. For the J-FIT DBA, η1 is calculated as follows:
where the parameter Tfixed_cycle represents the fixed polling cycle time. In the J-FIT DBA, for a given network load, η1 increases with the increase in fixed polling cycle time, because an increase in polling cycle time allows an ONU to spend more time in sleep mode. The percentage of energy-savings decreases with the increase in network load because Tactive increases with increased network loading, resulting in an increase in power consumption. The J-FIT DBA allows the ONUs enter into sleep mode for fixed polling cycle times ≥ 2ms. Due to the large difference in power consumption in sleep and active modes of a VCSEL ONU, the percentage of energy-savings of up to 65% could be achieved at low network loads with the J-FIT DBA algorithm. This is a significant improvement compared to the JIT DBA, where significant energy-savings are only possible at network loads ≥ 0.8. The J-FIT DBA therefore, results in improved energy-savings at a wider range of network loads.Figures 9(a) and 9(b) illustrate the average delay of a sleep/doze mode VCSEL ONU as a function of maximum and fixed polling cycle times and normalized network load of the JIT and J-FIT DBA algorithms, respectively. With the increase in polling cycle time, the average delay increases. For a given network load, with the increase in polling cycle time, the packets have to wait for a longer duration of time until they are allocated to be transmitted in the next transmission time slot. At a network load of 1, the link is fully utilized and there are more packets accumulated at each ONU queue leading to high queuing delays. The sharp increase in the average delay at the network load of 1 is a result of the unconstrained buffer length we have used in this algorithm. Compared to the JIT DBA, there is an increase in average delay at each network load for the J-FIT DBA. Nonetheless, when considering practical network loading levels of less than 0.6 [15], the average delay however remains below 1.32 ms and 16 ms for the JIT and the J-FIT DBA algorithms, respectively. Thus the delay values of both algorithms remain within an acceptable range of ≤ 100 ms [19] to support dominant and delay-sensitive services such as video over the 10G-EPON.
3.2. Performance comparison between VCSEL ONU and DFB ONU
Figure 10(a) illustrates the percentage of energy-savings achieved using a sleep/doze VCSEL ONU compared to an always-ON DFB ONU, as a function of maximum polling cycle time and normalized network load for the JIT DBA. The percentage of savings η2 is calculated as follows:
where the parameter PDFB,act represents the power consumption of an always-ON DFB ONU. As shown in the plot, at any given maximum polling cycle time, the sleep/doze mode VCSEL ONU results in energy-savings of 24% up to a network load of 0.8. Beyond the network load of 0.8, the sleep/doze mode VCSEL ONU achieves significant energy-savings with the maximum of 70% being reached at the network load of 1 and maximum polling cycle time of 10 ms. This increase in percentage of energy-savings results from the ONUs entering into doze mode up to a network load of 0.8 and into sleep mode beyond the network load of 0.8. As shown in Table 1, the difference in power consumption between an active DFB ONU (5.052 W) and a sleep mode VCSEL ONU (0.75 W) is higher than that of an active DFB ONU and a doze mode VCSEL ONU (3.85 W). This difference has resulted in higher energy-savings for network loads ≥ 0.8.Fig. 10 Percentage of energy-savings (η2) as a function of normalized network load and polling cycle time.
Figure 10(b) plots η2 as a function of fixed polling cycle time and normalized network load for the J-FIT DBA. Similar to η1, η2 also increases with the increase in fixed polling cycle time because the increase in fixed polling cycle time allows an ONU to stay in sleep or doze modes for a longer period. The percentage of energy-savings η2, decreases with the increase in network load as it keeps an ONU in active mode for a longer duration, resulting in an increase in power consumption. However, as explained previously, the J-FIT DBA algorithm results in low power consumption compared to the JIT DBA as J-FIT allows an ONU to enter into sleep mode for Tfixed_cycle ≥ 2ms. This results in larger η2 values in J-FIT across all the network loads, as compared to the JIT DBA. The results shown in Fig. 10 is an emphasis on the energy-efficiency of the sleep/doze mode VCSEL ONU compared to an always-ON DFB ONU. These results further strengthen the claim that the J-FIT DBA outperforms the JIT DBA in terms of energy-efficiency.
Figures 11(a) and 11(b) illustrate the percentage of energy-savings achieved using a sleep/doze VCSEL ONU compared to a sleep/doze DFB ONU, as a function of maximum and fixed polling cycle times and normalized network load of the JIT and J-FIT DBA algorithms, respectively. The percentage of savings η3 is calculated as follows:
where the parameter PDFB,doze/sleep represents the power consumption of the DFB ONU in sleep or doze modes. As the figures show, the sleep/doze VCSEL ONU outperforms the sleep/doze DFB ONU under both algorithms. As listed in Table 1, sleep/doze VCSEL ONU has a low doze-to-active transition time (330 ns) compared to a sleep/doze DFB ONU (760 ns). The sleep/doze VCSEL ONU also consumes less power in active mode, as compared to a sleep/doze DFB ONU. These characteristics of the sleep/doze VCSEL ONU has resulted in better performance in terms of energy-efficiency. It is also important to note that compared to the JIT DBA, the J-FIT DBA algorithm results in better energy-savings as J-FIT allows an ONU to enter into sleep mode for Tfixed_cycle ≥ 2ms. These results highlight the energy-efficiency of the sleep/doze VCSEL ONU and also of the J-FIT DBA.Fig. 11 Percentage of energy-savings (η3) as a function of normalized network load and polling cycle time.
4. Summary
In this work, we compared two dynamic bandwidth allocation algorithms designed to improve the energy-efficiency of the 10G-EPON. The algorithms, JIT DBA and J-FIT DBA, make use of the sleep and doze capabilities of a VCSEL ONU to reduce the energy consumption at the ONU when the ONU’s idle time ≤ 2 ms. The algorithms transit the ONUs into sleep or doze mode during its idle time and transit it back to active mode to receive data from the OLT just-in-time using MPCP GATE messages. The JIT DBA is designed with varying polling cycle times. The J-FIT DBA is designed with fixed polling cycle times to further improve the energy-efficiency at low network loads.
We have evaluated the performance of the algorithms with respect to polling cycle time, average power consumption, percentage of energy-savings, and average delay. We also analyzed the energy-efficiency of the sleep/doze mode VCSEL ONU compared to an always-ON DFB ONU under JIT and J-FIT DBA algorithms. Based on the simulation results, both JIT and J-FIT DBA algorithms result in improved energy-efficiency as compared to the always-ON VCSEL ONU. Further, the J-FIT DBA algorithm is shown to be much more energy-efficient across a wider range of network loads due to its fixed polling cycle time. In a further comparison between a sleep/doze mode VCSEL ONU and an always-ON DFB ONU, our results show that the VCSEL ONU outperforms the DFB ONU in terms of energy-efficiency. This is because the DFB ONU consumes more power in active mode, 5.052 W, in comparison to an active VCSEL ONU, 3.98 W. Introducing sleep and doze modes further reduce the average power consumption of the VCSEL ONU to 3.85 W and 0.75 W, respectively. Finally, the delay values of the JIT and the J-FIT DBA algorithms remain below 1.32 ms and 16 ms, respectively and well within the acceptable range of 100 ms to support the QoS requirements of practical access networks.
References and links
1. Y. Luo and N. Ansari, “Bandwidth allocation for multi-service access on EPONs,” IEEE Commun. Mag. 43(2), S16–S21 (2005) [CrossRef] .
2. J. Zhang and N. Ansari, “Toward energy-efficient 1G-EPON and 10G-EPON with sleep aware MAC control and scheduling,” IEEE Commun. Mag. 43(2), S33–S38 (2011) [CrossRef] .
3. M. Gupta and S. Singh, “Greening of Internet,” Proc. of ACM SIGCOMM , 19–26 (2003).
4. J. Baliga, R. Ayre, K. Hinton, W. V. Sorin, and R. S. Tucker, “Energy consumption in optical IP networks,” J. Lightwave Technol. 27(13), 2391–2401 (2009) [CrossRef] .
5. C. V. Praet, H. Chow, D. Suvakovic, D. V. Veen, A. Dupas, R. Boislaigue, R. Farah, M. F. Lau, J. Galaro, G. Qua, N. P. Anthapadmanabhan, G. Torfs, X. Yin, and P. Vetter, “Demonstration of low-power bit-interleaving TDM PON,” Opt. Express 20(26), B7–B14 (2012) [CrossRef] .
6. ITU-T G.Supp 45, “Means and impact of GPON power saving”.
7. T. Smith, R. S. Tucker, K. Hinton, and A. V. Tran, “Implications of sleep mode on activation and ranging protocols in PONs,” Proc. of 21st Annual meeting of the IEEE Lasers and Electro-Optics Society , (2008) [CrossRef] .
8. R. Kubo, J. Kani, Y. Fujimoto, N. Yoshimoto, and K. Kumozaki, “Sleep and adaptive rate control for power saving in 10G-EPON systems,” Proc. of IEEE Global Telecommunications Conference (GLOBECOM), (2009).
9. D. Ren, H. Li, and Y. Ji, “Power saving mechanism and performance analysis for 10 Gigabit class passive optical systems,” Proc. of 2nd IEEE International Conference on Network Infrastructure and Digital Content , 920–924 (2010).
10. S. W. Wong, She-Hwa Yen, P. Afshar, S. Yamasitha, and L. G. Kazovsky, “Demonstration of energy conserving TDM-PON with sleep mode ONU using fast clock recovery circuit,” Proc. of IEEE/OSA Opt. Fiber Commun. Conf. (OFC), OThW7 (2010).
11. IEEE 802.3ah, “Ethernet in the first mile task force,” (2004).
12. M. Fiammengo, A. Lindstrom, P. Monti, L. Wosinska, and B. Skubic, “Experimental evaluation of cyclic sleep with adaptable sleep period length for PON,” Proc. of 37th European Conference and Exhibition on Optical communication (ECOC), (2011).
13. M. P. I. Dias and E. Wong, “Energy-efficient dynamic bandwidth allocation algorithm for sleep/doze mode VC-SEL ONU,” Proc. of Asia Communications and Photonics conference (ACP), ATh1D.4 (2012).
14. M. P. I. Dias and E. Wong, “Performance evaluation of VCSEL ONU using energy-efficient just-in-time dynamic bandwidth allocation algorithm,” Proc. of Photonics Global Conference (PGC) , (2012) [CrossRef] .
15. E. Wong, M. Muller, P. I. Dias, C. A. Chan, and M. C. Amann, “Energy-efficiency of optical network units with vertical-cavity surface-emitting lasers,” Opt. Express 20(14), 14960–14970 (2012) [CrossRef] [PubMed] .
16. M. C. Amann, E. Wong, and M. Muller, “Energy-efficient high speed short cavity VCSELs,” Proc. of IEEE/OSA Opt. Fiber Commun. Conf. (OFC), OTh4F1 (2012).
17. “Cisco Visual Networking Index: Forecast and Methodology, 2011–2016.” Available: www.cisco.com.
18. E. Igawa, M. Nogami, and J. Nakagawa, “Symmetric 10G-EPON ONU BMT Employing Dynamic Power Save Control Circuit,”Proc. of IEEE/OSA Opt. Fiber Commun. Conf., Los Angeles, USA, NTuD5 (2011).
19. G. Kramer, B. Mukherjee, S. Dixit, Y. Ye, and R. Hirth, “Supporting differentiated classes of service in the Ethernet passive optical networks,” J. Opt. Netw. 1(8), 280–298 (2002).
